Download the paper: Virtualizing Recursion: Just-In-Time Graph Analytics in a Hyperscale Relational Warehouse
Graph analytics often pushes relational systems beyond the kinds of workloads they were originally designed to handle. Recursive queries can introduce complex execution patterns, large intermediate states, and difficult trade-offs between low latency and large-scale throughput.
This paper presents Ocient’s approach to recursive SQL and graph analytics using Dataflows, a procedural extension to SQL paired with a just-in-time virtualization runtime. The architecture automatically moves intermediate state between memory and NVMe as workloads change, enabling recursive and iterative graph algorithms to run directly within the relational warehouse.
The paper also examines use cases including bidirectional search, stable matching, Weakly Connected Components, and PageRank, along with performance results showing execution on graphs with more than 100 billion edges.